Tutorial: accessing the latest ESGF climate projections with Python

In parallel with the preparation of IPCC reports, there is another less visible project: the Coupled Model Intercomparison Project (CMIP), which coordinates the production of global climate projections. AR6, the 6th IPCC report published recently, corresponds to CMIP6, the 6th wave of the CMIP project.

These new projections offer higher resolution, better modeling of physical processes, new emissions scenarios, more SSPs than the RCPs from the previous IPCC report... In short, the CMIP6 projections are intended to rapidly replace those from the previous wave, CMIP5, published in 2013.

In this tutorial, we'll look at how to access these projections via the Earth System Grid Federation, the reference service for sharing climate projections.

First of all, what is a climate projection? And how can it be used?

A climate projection is a simulation of future weather. These simulations are carried out by Global Circulation Models (GCMs), which are very similar to conventional meteorological models.

There are many different climate models: some one hundred are contributing to CMIP6. The aim of the project is to coordinate the work, to ensure, for example, that projections made by NOAA's GFDL model in the USA are comparable with those of the Norwegian Climate Center's NorESM2 model.

Although general circulation models are cousins of meteorological models, a projection is not a forecast:

• a forecast aims to describe the state of the atmosphere at a given date,

• a projection describes one possible state, among others.

So how useful are climate projections? By taking a sufficient number of simulations, we can get an idea of the weather conditions that may prevail in the future.

As a general rule, 30 years are used. For example, if we want to estimate the average annual temperature in 2050, we can use a climate projection to calculate the average temperature for the years from 2036 to 2065. The distribution of these 30 simulations will give us a good idea of what's possible in 2050, and we could, for example, consider their median as the best estimate of temperature at that date.

Download projections with a browser via the ESGF websites

To put it simply, the Earth System Grid Federation is a peer-to-peer system for downloading climate projections. Via nodes hosted by various international research centers, you can find CMIP6 or CMIP5 projections, of course, as well as those from other major projects such as Cordex, and other lesser-known ones, such as Clim4energy, European projections for the energy sector.

These data can be accessed directly via a browser. To do so, click here:

• Go to the node of your choice. The screen that appears is slightly different for each node, but you should be able to click somewhere on CMIP6.

• Use the drop-down lists on the left of the screen to filter the projections you're interested in.

  For example, you can search for surface temperature (select “tas” in the Variable list) with a daily time step (“day” in Frequency) for the SSP2-4.5 scenario (“ssp245” in Experiment ID) with the IPSL model (“IPSL-CM6A-LR” in Source ID).

• Once you've made your selection, click on “Search”. At the time of writing, 11 results have been obtained, corresponding to variants with slightly different initial conditions, for example.

• Take the first variant (r1i1p1f1), click on “List Files” in the corresponding line and you'll get a list of the files making up this projection. In this case, there's only one.

• All you have to do is click on “HTTP Download” to start the download.

This may seem a practical way of doing things, but in reality it's not very suitable: in a real project, you're always using several emission scenarios and several models, often with several variables... which quickly adds up to dozens of projections, or even hundreds for more complex studies...

So, manual downloading is out. In any case, you'll have to use a programming language such as Python, R or MATLAB to open and use them. So why not use it for downloading too?

Access CMIP6 projections with Python

It's very easy to access CMIP6 projections from ESGF with Python. There's a package for this: ESGF PyClient.

Let's start by installing it:

conda install -c conda-forge esgf-pyclient

We will then create a connection to the IPSL node:

“distrib = False” means that the search will be performed on this node only. If this parameter is set to True, the search is performed on all nodes, which can lead to more exhaustive results, but also to duplicates.

From this connection, we can search for the projections we're interested in. As before, we will search for :

• CMIP6 projections

• for the IPSL model

• for the SSP2-4.5 scenario

• for surface air temperature

• with a daily time step

recherche.hit_count allows you to see how many results our search returns. In this case, as with the manual search, we get 11 corresponding to different variants.

We can display the list of these projections :

As before, we're going to download the first projection, corresponding to the r1i1p1f1 variant.

The urls of the corresponding files can be obtained as follows:

Once you have these urls, the projection can be downloaded and opened directly with Xarray (for more information on this library and examples of use, see our previous tutorials):

What's next?

It is now possible to use this projection. For example, we can calculate annual temperatures on the American continent, their average at the end of the century and changes between 2021-2050 and 2071-2100 :

Then plot the result on a map:

Here is the resulting map:

It's immediately obvious that spatial resolution is limited. This is one of the reasons why these projections are never used as they stand.

In a real project, there would still be several stages to go through before obtaining a usable result:

1. Debiasing: climate models are generally biased, so before a projection can be used, it must be cleansed of these systematic errors.

2. Downscaling: this involves improving the spatial resolution of projections using a regional climate model or statistical methods.

3. A calculation stage: in general, we're not looking for a simple temperature series or projection, we need to calculate the desired indicator: changes in monthly, seasonal or annual averages, changes in extremes, or the assessment of the concrete effects of climate change at local level, for example on the intensity and frequency of heatwaves or on thermal comfort.

4. Evaluating multi-model uncertainties: There are many climate models out there, and they can give divergent results. So it's important to avoid using just one model, and instead rely on a set of models suited to the study to be carried out.

As you can see, access to projections is only the first step. There's still a lot of work to be done to produce results that are in line with the state of the art and useful for real economic or political decisions.

---

👋 If you've come this far, it's probably because you really do have a project that requires climate projections... So let's face it: you can try to carry out these studies yourself, but in the vast majority of cases calling on Callendar will be quicker and cheaper. Contact us to discuss your project!

Callendar specializes in developing affordable solutions for local climate risk assessment. We have already helped numerous companies and over 100,000 individuals to anticipate the effects of climate change.